Generative AI got all the headlines last year. Now, something bigger may be brewing — and it’s not just about writing essays or generating fancy art. At Nvidia’s GTC conference, CEO Jensen Huang declared that the emergence of AI agents represented “a multi-trillion-dollar opportunity.”
His point? We’re moving past artificial intelligence (AI) that creates content and into a world where AI agents are actually handling tasks, automating systems and behaving more like coworkers.
“The IT department of every company,” he joked, “is going to be the HR department of AI agents in the future.”
So, how are AI agents different from what we’ve seen before? Which companies are pushing it forward, and how might this reshape the way businesses — and individuals — operate? Let’s break it down: Who’s leading, where it’s showing up in real-world use cases, and what you need to know before letting a digital agent take the reins.
What is an AI agent?
An AI agent is sophisticated software that carries out tasks. It can respond to its surroundings, learn, and use other software tools.
“Unlike a chatbot where you type in prompts and you get responses, an AI agent is autonomous and can think and act on its own accord,” said Aaron Chaisson, VP of product and solutions at VAST, a data platform company involved with AI. “It will plan and carry out multiple steps to achieve goals.”
He provided an example: “Suppose you have a Visa card and go to the company’s website. You say that you want to take a trip to London next week. One AI agent will book the flights and another will book the hotels. Then another agent will provide recommendations for restaurants and sites to visit.”
Note, this is not to imply that AI agents are completely autonomous. There will still be some queries and requests for approvals. Even so, AI agents promise to greatly streamline tedious processes.
The best AI agents
Determining the best AI agents is challenging, as the field is still in its early stages.
“AI agents aren’t ready for production in most use cases,” Andy Triedman, a partner at Theory Ventures, which invests in software companies, told Kiplinger. “Many tasks are just too complex for current models, and even for ones that are possible, models won’t be successful 100% of the time.”
So, there continues to be ongoing progress and innovation. Part of this work is being done by AI model developers like OpenAI, Anthropic, Microsoft, Amazon and Meta, as their systems are becoming increasingly agentic.
Consider OpenAI’s Operator platform. It can visit websites, fill out forms, place orders and schedule appointments. Currently, Operator is in research preview and only available for the Pro subscription, which costs $200 per month per user. But OpenAI has plans to make it a feature of its other plans.
Meanwhile, large enterprise software firms – like Salesforce, ServiceNow, Oracle and Microsoft – are developing AI agents to automate business operations. A key advantage is that these companies have access to enormous amounts of data and workflows, which allows for training AI agents to better handle automations.
Take Salesforce’s Agentforce platform, which makes it easy to build and deploy AI agents (in some cases, the process can only take a few minutes). The company started producing television commercials for it featuring actor Matthew McConaughey, often alongside Woody Harrelson, signaling the product's strategic importance.
“Agentforce goes beyond chatbots and copilots, using advanced reasoning abilities to make decisions and take action, like resolving customer cases,” Kishan Chetan, EVP & GM of Service Cloud at Salesforce, told Kiplinger.
“For example," Chetan continued, "the Sales Coach Agent can help coach salespeople, letting them practice their pitching, how they respond to objections, and how to negotiate deals with realistic role-playing that is tailored to each deal for every seller. Agentforce can even provide feedback on seller strengths, reveal areas for improvement, and arm them with actionable steps to help them advance deals, all grounded in the business’s CRM data.”
There are also plenty of startups creating interesting AI agents:
- MindCorp: Michelle Crames, the founder and president of this startup, was previously a business analyst at McKinsey & Company. She saw how time-intensive the management consulting process was, so she launched MindCorp to use AI agents to significantly improve the process. A project that may take a couple weeks could instead be done in a few hours.
- Sierra: The co-CEO and founder is Bret Taylor, who was the chief technology officer of Facebook and co-CEO of Salesforce. As for Sierra, it uses AI agents for customer service and support, with customers like SiriusXM, SONOS and ADT. Late last year, the company raised $175 million at a $4.5 billion valuation.
- Hippocratic AI: This startup uses its own AI models to create agents for health care tasks like pre-operative check-ins, post-discharge follow-ups, chronic care management and appointment scheduling. In January, the company announced a $141 funding round at a $1.64 billion valuation.
The risks of AI agents
The autonomous decision-making of AI agents poses major potential risks. For example, a system could misinterpret data and approve a large invoice. Or, a health care AI agent might misdiagnose conditions due to biases in training data, which could lead to inappropriate treatments.
“AI agents need to be constantly updated with the latest data for both accurate outputs and to prevent false feedback loops that may put the organization at risk of interruptions to managing day-to-day tasks,” said Chris Moss, general manager of small business solutions at Dun & Bradstreet. “It’s crucial to still have a human in the loop to ensure there is proper oversight.”
If you're experimenting yourself, you should always be sure to protect your privacy while using AI by keeping up with the latest security information and checking where exactly you're sharing information and how it's used.
Another major issue is that today’s information technology (IT) infrastructures were not built for the complexities of AI agents.
“AI agents need to be able to integrate and authenticate with different applications,” said Triedman. “They need a secure environment to run code. And they need observability and analytics to help developers understand where they’re going wrong. Companies are rapidly working on these problems, but we don’t have all the solutions yet.”
